Gemini in Google Sheets: AI Chart Generation Tested

Every spreadsheet vendor is now promising the same thing: describe the chart you want in plain English, and the AI builds it. Google’s version is Gemini in Sheets, and the marketing is confident — “turn raw data into charts and graphs quickly, reducing the need to manually build scatter plots, bar graphs, pie charts, and more.” That’s a real claim about a real workflow, so it deserves a real test rather than a rewrite of the press release.

The short version: Gemini genuinely builds charts from a prompt, and for straightforward asks it’s faster than clicking through the Chart editor. But “faster than the menu” is a low bar, the output is less flexible than the marketing implies, and there’s one limitation that will annoy anyone who wants a chart they can keep editing. Here’s what actually happens when you use it.

What Google says it does

Worth being precise about the claims, because Google has shipped several waves of this and the naming blurs together.

The chart-generation capability arrived in February 2025, when Google added Gemini-powered visualisation to Sheets — the ability to generate insights like correlations and trends and to produce charts, including heatmaps inserted as static images over cells. A broader Workspace update in March 2026 turned Gemini from a sidebar novelty into something that pulls context from Gmail and Drive. Then on 22 April 2026 came the headline feature Google calls “end-to-end creation” — Gemini building and populating entire spreadsheets, dashboards included, from a single conversational prompt.

Google also did something most vendors don’t: it published a third-party benchmark. On SpreadsheetBench, an independent public test of spreadsheet manipulation, Gemini in Sheets scored 70.48%. That number is more meaningful than an internal benchmark precisely because Google didn’t build the test. It’s a genuine data point. It is not, however, a claim about charts specifically — SpreadsheetBench measures spreadsheet tasks broadly, and a 70% task-success rate also means roughly three in ten attempts miss. Keep that ratio in mind.

How chart generation actually works

You open the Ask Gemini panel from the icon at the top right of the sheet, then type what you want: “Create a chart showing revenue by region,” or “Chart the trend in column C over the last 12 months.”

Gemini analyses the data and returns a chart in the side panel — not on the sheet. You then click Insert to drop it onto the grid where your cursor sits. Two useful touches: an Analysis steps link shows the reasoning it followed, and a Sources link appears when it pulled from a Drive file or Gmail thread. That transparency is a genuine point in its favour — you can check its working rather than trusting a black box.

One requirement the documentation is honest about, and that trips people up: this works best on native Google Sheets files. If you’re in an uploaded .xlsx, you’re told to convert it via File > Save as Google Sheets first. Skip that and the feature underperforms or won’t engage.

Where it performs well

Tested against the kind of requests real people make, Gemini is most reliable at the boring, common stuff — which, to be fair, is most of what anyone charts.

Clean, well-labelled data with an obvious chart. Give it a table with clear headers — “Month,” “Sales,” “Region” — and ask for a standard bar or line chart, and it delivers a sensible result quickly. Google’s own guidance confirms the dependency: chart creation works best when your data has headers for each column or row. That’s not a limitation so much as a tell — the AI leans heavily on your column names to understand intent.

“What’s the trend” questions. Ask it to identify and chart a trend and it does a competent job of picking a time-series line and highlighting the direction. For someone who isn’t sure which chart fits, this removes a decision. Whether it picks the right chart is a separate question — see below.

First drafts of a dashboard. The April 2026 end-to-end feature will assemble a populated dashboard from a prompt. As a starting point that saves the blank-page problem, it’s legitimately useful. As a finished deliverable, it isn’t — you’ll be editing.

Where it falls short

This is where the marketing and the reality separate.

Chart choice is competent, not expert. Gemini reliably produces a chart. It does not reliably produce the best chart. Ask it to visualise part-to-whole data and it will happily hand you a pie chart with eight slices — technically a chart, genuinely a poor choice. The AI optimises for “a valid visualisation of this data,” not “the visualisation that communicates this honestly.” It has no opinion, and charts need opinions. If you don’t already know which chart you should be using, Gemini won’t reliably teach you — it’ll just confidently build whatever’s plausible.

The static-image heatmap problem. Those heatmaps Google highlights? They insert as static images over cells, not as live, editable chart objects. That means no recolouring, no re-ranging, no updating when the underlying numbers change. You get a picture. For a one-off screenshot that’s fine; for anything you’ll revisit, it’s a dead end, and the marketing doesn’t lead with that caveat.

It’s a side-panel result, not a native chart — until you insert it. Because the output lands in the panel first, the iterative loop is clunkier than it sounds. Refining (“make the bars horizontal,” “sort descending”) means re-prompting rather than nudging a chart you can see, and the results drift. For fine control you’ll end up inserting the chart and then finishing it by hand in the ordinary Chart editor anyway — at which point Gemini saved you the creation step but not the refinement step.

The 70% ceiling shows up as inconsistency. That benchmark figure isn’t abstract. Run the same prompt on similar data twice and you won’t always get the same chart. For casual use that’s tolerable. For anything where you need a repeatable, defensible output, the variability is a real cost — you have to verify every result rather than trusting the tool, which eats into the time it supposedly saves.

The honest verdict

Gemini’s chart generation is a real feature that does real work — this isn’t vapourware, and the published benchmark keeps it honest in a way most vendor claims aren’t. For a professional who lives in Sheets and wants a quick standard chart from clean data, it removes clicks and, occasionally, an entire dashboard’s worth of setup.

But it’s an accelerator, not a replacement for knowing what you’re doing. It builds valid charts, not good ones; its best output still needs a human editing pass; and its most-marketed trick (heatmaps) ships as static images you can’t touch. The right mental model is a fast junior analyst who’s technically capable but has no design judgement and occasionally gets it wrong — useful, supervised, never trusted blindly.

Set against Microsoft’s equivalent, the two are converging on the same pitch. If you’re weighing the ecosystems, our guide to Copilot charts in Excel covers Microsoft’s side of this race, and our Excel vs Google Sheets charts comparison puts the two spreadsheet platforms head to head on visualisation specifically.

If you’d rather skip the AI lottery entirely and start from something already built correctly, our Google Sheets chart templates give you copy-to-Drive charts with the design decisions already made — which, for a lot of tasks, is still the faster and more reliable route.

FAQ

Do I need a paid plan to use Gemini chart generation in Sheets? Yes. Gemini in Sheets requires an eligible Google Workspace or Google AI plan; premium Gemini features are now bundled into Workspace plans rather than sold as a separate add-on. Personal-account users may access it through Google’s Workspace Experiments trusted-tester programme.

Why won’t Gemini generate a chart from my Excel file? The feature works best with native Sheets files. Open your .xlsx and use File > Save as Google Sheets to convert it first, then try the prompt again.

Can Gemini pick the right chart type for me? It picks a valid one, not necessarily the best one. It has no design judgement — it will produce a cluttered pie chart if that technically fits your data. Decide the chart type yourself, then let Gemini build it.

Are the charts Gemini makes editable afterwards? Standard charts you insert onto the sheet are editable in the normal Chart editor. But heatmaps insert as static images over cells and can’t be edited or updated — a limitation the marketing tends to skip.

Is Gemini in Sheets better than Copilot in Excel for charts? They’re broadly comparable and improving fast. Both build valid charts from prompts and both need a human finishing pass. The better choice usually comes down to which ecosystem you already work in — see our Excel-vs-Sheets comparison for the specifics.

Feature availability and behaviour described here reflect Google’s announcements through mid-2026. Google is iterating on these tools rapidly, so specific capabilities may have changed since publication — verify current behaviour in your own account.